WHERE to Generate Matters: Budget-Aware Synthetic Augmentation for Label Skewed Federated Learning

2026-07-07 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

FedEAS is a novel policy designed to address label skew in federated learning (FL), a common issue causing client drift and global accuracy degradation. While synthetic data augmentation can mitigate this imbalance, full class balancing incurs substantial computational costs. FedEAS introduces an entropy-adaptive per-class generation budget for each client, which intelligently determines both the quantity of synthetic samples to generate and their distribution ("WHERE" they go). This approach allows the total generation budget to be dynamically derived from per-client needs, rather than being fixed. FedEAS recovers most of the accuracy gains achieved by full class balancing, yet reduces the generation budget by 94.1%. Furthermore, it outperforms Uniform allocation by up to 18.82% on CIFAR-10 and CIFAR-100 datasets at the same total generation budget.

Key takeaway

For Machine Learning Engineers developing federated learning systems with label skew, you should consider implementing budget-aware synthetic augmentation policies like FedEAS. This approach significantly reduces computational costs associated with full class balancing while maintaining high accuracy. By dynamically allocating generation budgets based on client-specific label distributions, you can optimize resource utilization and improve model performance across diverse client environments.

Key insights

Budget-aware synthetic augmentation using entropy-adaptive per-class generation effectively mitigates label skew in federated learning.

Principles

• Label skew degrades global accuracy in federated learning.
• Synthetic data augmentation can reduce label imbalance.
• Full class balancing is computationally expensive.

Method

FedEAS assigns clients an entropy-adaptive per-class generation budget based on local label distribution, determining both generation quantity and sample destination.

In practice

• Reduce FL computation cost by 94.1%.
• Improve accuracy by up to 18.82% over Uniform allocation.

Topics

• Federated Learning
• Label Skew
• Synthetic Data Augmentation
• Budget Optimization
• FedEAS Policy
• Client Drift

Best for: Research Scientist, AI Scientist, Machine Learning Engineer

Related on AIssential

• FedSIR: Spectral Client Identification and Relabeling for Federated Learning with Noisy Labels — FedSIR uses spectral analysis of client features to identify and relabel noisy data in federated learning.
• Generative Augmented Inference — GAI consistently improves estimation efficiency and reduces human labeling needs by integrating AI outputs as informative features.
• Amortizing Federated Adaptation: Hypernetwork Driven LoRA for Personalized Foundation Models — HyperLoRA uses learned operators for federated adaptation, overcoming LoRA's aggregation bias and client initialization lag.
• Metric-Dependent Annotation Saturation for Learning from Label Distributions — Annotation budget needs depend on the target evaluation metric, as different metrics saturate at varying annotator counts.
• In-Context Multiple Instance Learning — Pretraining a Perceiver-style in-context learner on synthetic MIL data enables few-shot classification without gradient updates.
• Federated Nested Learning: Collaborative Training of Self-Referential Memories for Test-Time Adaptation — FedNL enables federated models to learn adaptive memory update rules, not just static weights, for robust test-time adaptation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.